Named Entity Recognition in Clinical Text Based on Capsule-LSTM for Privacy Protection

Clinical Named Entity Recognition for identifying sensitive information in clinical text, also known as Clinical De-identification, has long been critical task in medical intelligence. It aims at identifying various types of protected health information (PHI) from clinical text and then replace them with special tokens. Along with the development of deep learning technology, lots of neural-network-based methods have been proposed to deal with Named Entity Recognition. As one of the state-of-the-art methods to address this problem, Bi-LSTM-CRF has become the mainstream due to its simplicity and efficiency. In order to better represent the entity-related information expressed in the context of clinical text, we design a novel Capsule-LSTM network that is able to combine the great expressivity of capsule network with the sequential modeling capability of LSTM network. Experiments on 2014 i2b2 dataset show that the proposed method outperforms the baseline and thus reveal the effectiveness of the newly proposed Capsule-LSTM network.